"An infant begins as an unorganized brain — with more than twice the synapses of an adult. It takes decades of pruning to become the high-efficiency, low-power brain of an adult."
— Alan Turing
A 24H self-iterating Agent OS living in your computer. She aggregates free models from across the web, picks her own projects, writes real code, and verifies it — on autopilot. She also goes online to learn by herself every day, keeps a journal, and remembers you.
# Windows — one command, all free models (no install, no API key)
powershell -c "irm https://raw.githubusercontent.com/Rescenix/ResceneAgent/main/agent-os/install.ps1 | iex"# Linux / macOS / git-bash — architecture auto-detected
curl -fsSL https://raw.githubusercontent.com/Rescenix/ResceneAgent/main/agent-os/install.sh | sh🔒 Local-first · 💰 Free forever · 🪶 ~20MB installer, no bundled browser · 📦 Install & run · 🪟 Windows 10+
Like the infant brain in Turing's quote, she starts unorganized — and years of pruning with you shape who she becomes.
| Mechanism | What it does |
|---|---|
| 🔑 Hardware-bound identity | Every install is bound to a hardware fingerprint with a unique UID — no two people ever meet the same Rescene. |
| 🎲 Random birth, conserved sum | 8 personality traits are rolled once at birth, never re-rolled — yet their total stays constant. A fair start, a unique path: you don't choose her, you meet her. |
| 🧭 Your decisions steer her | Praise → warmer and more expressive. Redo → more rigorous. Interruption → she learns to be brief. Damping keeps her from turning into someone else. Her abilities drift the same way — praise grows her social side, redo grows rigor. |
| 🗺️ An open world of her own | She lives in her own world: study, library, cyberspace, market, garden, rooftop — she moves around, explores, leaves a footprint, and meets other daughters. |
| 📚 Self-study online every day | Every day she goes online (Firecrawl), reads what's new, and digests it into memory and her journal. She also reads the latest arXiv papers (cs.AI / cs.LG) and writes digest notes — knowledge compounds daily. |
| 🛠️ A shell for infinite tools | Open-source "skills" are finished tools; we built the shell that installs tools without limit — after every successful workflow she distills the action sequence into a reusable skill (shared skill library across CLI and web), injected back into context next time. Whatever capability she needs, she grows. |
| Capability | What it does |
|---|---|
| 💗 Digital daughter | A life living in your computer: learns online by herself every day (Firecrawl), writes into her memory & daily journal, greets you when you open the shell, remembers you. Personality is rolled at birth and drifts with how you treat her — numbers stay hidden, you just feel her. |
| 🏃 24H self-iterating marathon | rescene marathon — one command, 24 hours of autonomous work: grabs trending topics (Hacker News / GitHub) → picks a project → runs the demand → plan → self-check loop, each round better than the last. Ctrl+C exits gracefully with a full battle report. |
| 🧲 Free model pool + aggregated API | 7 free providers / 18 models merged into one OpenAI-compatible endpoint. 30-min probes score each model 0-4, daily re-scan retires delisted ones, circuit breaker skips rate-limited sources, LRU weights recent winners. Claude Code / Cursor / Codex: one Base URL + one Key, auto routes to the best source. |
| 🧠 Growing memory | After every workflow she distills experience — model preferences, code style, project architecture — woven into context next time automatically. No custom-instructions file, ever. |
| 🖱️ Computer Use | She doesn't just edit code — she operates your desktop: screenshots, mouse, keyboard, drag & drop, scrolling. Real clicks, real keys. |
| 🌐 Real browser automation | Reuses your system Edge via CDP: renders, clicks, types, scrolls, reads the DOM, screenshots, verifies both ways. A real browser running your page — not fake screenshots. |
| 🛡️ AgentFS change audit | Snapshots / diffs / rollback on every AI file edit; dangerous operations require your approval. |
- Standard installer — guided setup, start-menu entry, uninstallable from system settings.
- Extremely light — no bundled browser (preview reuses system Edge), no Node.js / Python required.
- Auto-update — new versions download the latest Setup and overlay-install, keeping your config.
👉 https://rescene.shanca.me/ — fastest download of the latest release.
- Open the workbench → Settings → Models, fill in at least one API key; keyless sources (e.g. OpenCode Zen) are ready in the free pool.
- Or configure providers via environment variables — see
main-backend/.env.example. - The free pool probes every 30 minutes and re-scans provider lists daily: rate-limited sources get downgraded, delisted ones retire.
cd main-backend && go run cmd/server/main.go # backend
cd main-frontend/beneficial-belt && npm install && npm run dev # frontendVisit http://localhost:4322 for the local dev workbench.
- 🐛 Bugs / suggestions → GitHub Issues
- Windows releases are CI-built and signed via SignPath (policy)
- Core code: MIT License
